Deepfake detection tools don’t always work. Here’s proof | usagoldmines.com

By now, you know what a deepfake is. These AI-generated audio and video clips keep spreading, causing louder and louder warnings about them every week.

It’s easy to mistake these imitations of people for the real thing. (Heck, I wrote a confession not long ago about failing a deepfake video quiz.) Security tools have popped up to help, with established software vendors and newer companies alike making deepfake detectors available. But the sad news is… they’re not as reliable as you might hope.

Look at this screenshot from a TikTok video, highlighted by Instagram user natalyamakes:

TikTok

As natalyamakes points out, the video seems innocent at first… right up until the person jams her fingers through her nose. (You know, like any normal crying human who reaches up to wipe their nose and goes right through the cartilage separating their nostrils.)

It’s an incredible video. But it’ll pass scam detectors.

I tried several publicly available ones—the top hits on Google search. Links that people would likely try when making an effort to determine if what they’re seeing is real. And, after multiple passes, some scanners could not give me consistent estimates… nor very accurate ones.

For example, deepfakedetector.ai said that the probability of the video being a deepfake was anywhere from five to 24 percent. The reasoning? The video had clear facial details and expressions, convincing eye movement and blink rate, consistent lighting, natural interaction with objects, realistic details and depth of focus, and subtle motion blur.

Deepfakedetector.ai

(Again, the subject of the video put her fingers through her nose cartilage.)

That’s not to say that all such tools are unreliable—Hive Moderation’s detection demo pegged this clip as being AI-generated, for example.

But you know what also works? A pair of eyes and attention to detail.

As always, a healthy sense of skepticism helps with survival on the internet. This particular scheme perpetuates fraud, but a more benign version—as natalyamakes notes, the account creator likely is a dropshipper attempting to portray mass-produced goods as handmade, deserving of a higher price tag.

You can nip that in the bud and avoid overpaying for goods or getting outright swindled for fake or nonexistent products by grabbing a screenshot, then doing a quick reverse image search on Google. You can also search for keywords or phrases describing the item. Not only will you sniff out fake handmade goods, but you can end up saving money when doing regular shopping with this trick, too.